The "fortified essays" seem quite interesting, and reading the description, it doesn't seem like it's a full-time job, so maybe I'll do it on the side. I could imagine writing things in the style of "X: much more than you wanted to know"
I’m writing to lodge a minor objection to the title of this series. I find it hard to believe I’m the only one who, every time one of these comes in, finds himself stuck with that horrible Bangles song from the 1980s stuck in my head for the rest of the day. (“Just another man[t]ic Monday.”)
For anyone too young to know the song—for the love of God, don’t look it up, at least not while this series continues! Much like Roko’s Basilisk, you will rue the weekly suffering you endure due to knowledge that you can’t unlearn!
I already struggle with this problem whenever I have the thought "huh it's Monday" and Scott is certainly not helping. I wouldn't mind, except "wish it was Sunday / 'cause that's my fun day" has always grated for some reason.
Ummm ... that song was actually written by ... Prince.
It's also, not by coincidence, really, REALLY good. Does it stick in your head? Yes. Prince was an amazing pop songwriter.
Knowing that it's actually Prince...
(who was smitten with Susanna Hoffs, but let the Bangles go ahead and record MM even though she [not unkindly] turned him down)
... go back and listen to the bridge and see if you now smack yourself in the forehead, that bridge is so clearly his genius...
This was during his absurd imperial phase, when his output was so great that even his table scraps could be top 10 hits and he launched bands seemingly every few months.
Why should Kalshi be less than 10% on their fees? That’s what a typical sports book would be charging, and their motivations (i.e. profit) are the same. I guess a few online sports books go cheaper, but they have more competition. Frankly I’m surprised Kalshi’s fees are that low if they’re the first regulated company into this market.
Ignoring the issue of "should" (since you're right, it's competition driven), the comparison to sportsbooks isn't quite right. While sportsbooks do resemble a market at times as they try and balance their action, they can leave themselves substantially exposed to risk on one position or another.
Kalashi doesn't take on any risk however. They're operating an exchange without taking on any risk, so in the abstract, you'd expect their fees to be lower.
I agree that they should be lower! But I'm very confused bc I think they are lower. The PDF Scott linked shows the largest fee as $0.0175 for $0.50 of profit, so isn't that 3.5%?
I think neither of these are fair comparisons, since both make their profits heavily through other avenues that aren’t open to Kalshi.
First off, a lot of modern companies charging 0% royalties make money in other ways, the primary ones seem to be matching you with other customers to cash in on spreads, selling the right to place your order (I don’t understand why this works) and micro-interest from when your money sits around between transactions. Assumedly since Kalshi is the market there aren’t spreads (so you’re saving money here), they place orders themselves, and their volume is low.
With credit cards, they do charge a percentage amount, but they charge it to the vendor (though I don’t think it’s 10%). But the way they make tons of money is off of the ~20% APR loans customers sometimes take out.
I looked it up, credit cards range from 1.5% to 3.3% which at the high end is comparable to Kalshi if you take the 3.5% rate the representative said it evened out to.
My understanding is that this only holds for users with good credit. I’d had friends who were unable to get cash back cards (or even who had to get secured cards) because their credit was insufficient.
considering that credit scores come from how much (and old) lines of. credit you have, how much of it you use, and how reliably you pay bills, this all adds up to “how much do you actually need a credit card?”
Not having a credit card would literally not affect my day-to-day life. Thus they have to compete to convince a customer like me to actually get a card. So they give me rewards like cash back. For someone who actually needs the card, they’ll usually have worse credit and thus not get as good a card.
So in my view, what’s going on is that Kalshi hasn’t figured out a model for price discrimination yet. I expect if we ever have major traders on such a company, they’ll be able to negotiate reduced rates.
Having done some research on the payments industry, at least in the US, I think credit card fees are seldom higher than 2.0% these days (unless we're talking PayPal). And they've been trending consistently downward.
The loans are made by the banks that issue cards, and for the banks I think those are usually the most lucrative part of the credit card business, but the banks aren't the only players. The "networks" like Visa and Mastercard take the smallest part of the fee (I think usually less than 0.5%), but they're very profitable because they also do the least work. IIRC the payment processors (e.g. Global Payments) take the largest part of the fee, or perhaps they're roughly tied with the banks, but they haven't been as profitable because they're also out there doing the most work and they have the least market power.
All this is different for AMEX, as they're vertically integrated and own the whole process. But they're also much smaller.
There should be a "things that make vaccines fail" prediction. Turns out, steroids +/- 2 weeks from vaccination can lower the effectiveness. Since shots are being given by big-box stores and not updating to PCP charts, there is a likelihood of this type of thing. It may be contributing to me getting breakthrough covid, fortunately not a bad case, but annoying, because not supposed to be happening. How about, percent of breakthrough infections influenced by drug interactions w/vaccine. If I can figure out how to work it I'll put that up.
"percent of Americans who die of COVID between now and 1/1/22 who were fully vaccinated" is ambiguous, and I first read it as "X% of the set of Americans will have been vaccinated and also die from COVID" rather than "out of the set of Americans who die of COVID, X% will have been vaccinated". Thus my WTF reaction to your estimate of 2.5%.
I don't think the original wording is actually ambiguous, although I also misunderstood it on a first reading. (Probably because I was expecting a different question.)
Oh, wow, I had the same misreading. Probably because the number I am most interested in is "as a fully vaccinated person, what percent chance do I have of dying/hospitalization/long COVID?" so I just assumed that paragraph was about the question I had in mind.
(As for the answer to my question: all the articles seem to say "almost no chance of dying/hospitalization", which is reassuring, although as Scott mentions having an actual predicted number would be better. The articles I've seen are quiet on long COVID speculation though, besides a throwaway line about "concern among scientists" in https://www.nytimes.com/2021/07/22/health/coronavirus-breakthrough-infections-delta.html .)
Is there any long-term prediction markets covering ambiguous questions in the realm of science? I'd be particularly interested in what they would predict the field of nutrition science will say about e.g. what is the healthiest way to eat in 10, 20, 30 years. Will they have settled on low-carb? Back to low-fat? Etc.
Doesn't that just mean changes to the retirement age above 64 won't affect these numbers? The retirement age for most people in China is currently below 64 (at least officially - I don't have any idea how common it is for people to work beyond the official retirement age). Sorry if I still don't understand how this actually works.
No, it won't affect the numbers. Lots of developing countries don't even have retirement ages, so a fixed cutoff is the simplest way to quantify who is too old or too young to work.
China's median age is around 37, so their demographic crisis won't hit till after 2030 (around 2040 I guess) when the effects of the one-child policy will become obvious in the workforce. Unless they open to mass immigration of course.
China is pushing now for women to have more children - they have just instituted their Third Child Policy. So between social pressure (and China already has its social credit ratings system in place) to stop being a career woman, get married, and start a family, and already married people who might want more kids, presumably the estimation here is that China will be getting a lot of new babies over the years.
I have also read online some rather ugly speculation that this is also part of the treatment of the Ugyhurs: move in Han men, get them married (consent not being too important) to Ugyhur women, and start producing bouncing Han Chinese babies which is two birds with one stone about propping up the birth rate and 'assimilating' the Ughyur minority into the mainstream. How true this is I have no idea, but I can't dismiss it out of hand.
The cutoff is in fact 19- and 65+ for the metaculus question. So every child that is born this year or later will *increase* the dependency ratio in 2039, not *decrease*. This is the wrong direction.
I'm skeptical that raising the allowed number of children to 3 will have that much impact on the birth rate. I don't think there was that much of a bump when they went from 1 to 2. There's already a lot of social pressure in China to have a "traditional family" but the default is 1 kid and not many people actually want / can afford more.
China is certainly trying to find ways to alleviate the pressures that deter people from having more kids - that's why they're cracking down on these after school tutoring companies right now. It just doesn't seem to be very easy to get people to have more kids though (see for example Singapore).
This is right. Raising fertility in modern societies is extremely difficult, the anti-natal pressures within modernity are immense, and they seem to affect NE Asian societies particularly acutely. China maybe has a chance of pushing back, but they haven't done anything yet to appear very serious on this topic.
"Appearing serious" would probably involve things like promoting anti-natal media and banning all media that so much as hints at glamorizing childlessness or small families, and requiring 3+ children (or even more) for promotion within the CCP.
I also think that if these things worked, it might still take decades. Pro-natal propaganda probably won't have much effect on people who are 25 today, but 20 years of it might affect people who are 5 today.
Scott, where did you get the 10% number from for the Kalshi fees? The document you linked shows the largest fee as $0.0175 for $0.50 of profit, so isn't that 3.5%?
So I'm trying to see what markets are on Kalshi but I can't find a page for it (without signing up, anyway -- I just want to see all the markets, I don't want to actually sign up). Is there a way to see this? I can find pages for individual markets, but not for the overall list of markets. (It isn't https://kalshi.com/markets/ ; that doesn't work.)
One ought to be able to see the list of markets without signing up!
Am I misunderstanding or are the fees for Kalshi quite a bit less than 10%?
It looks like for 1-cent contracts the fee is 7%, but it looks like it is lower than that for higher-cost contracts - e.g. the fee for one hundred 50-cent contracts is $1.75, which is only 3.5%.
And for one hundred 99-cent contracts, the fee is 7 cents, which is less than 0.1%.
So on average the fee is around 3.5%? I might be misunderstanding the fee schedule though.
> We charge this fee only to “liquidity takers,” who place orders that match orders sitting on the orderbook to complete a trade. We do not charge a fee to “liquidity makers” who place an order that rests on the orderbook waiting to be matched.
So it seems the fee is only for one side of a trade. So on average your fee would be half of that?
I've been trading on Polymarket for two months, and it's been incredible. I am learning a lot more on the subjects I am wagering on, and it is still vastly inefficient. I have seen some people do very little work to guarantee themselves 5 figure payouts. This is what trading bonds in the 80s much have felt like to an extent.
I should note that I just put $100 into this and I had to pay for almost $30 of Etherium on top of the USDC to get my $100 in, so be aware. (My $100 was $99.64 after some fees but that's the least of it.)
I had not heard of dependency ratios before, and I was VERY surprised at those numbers until I looked "dependency ratio" up on Wikipedia and discovered that it is conventionally multiplied by 100 (i.e. expressed as a percentage). So a dependency ratio of "50" means 0.5 dependents per working-age person, not 50 dependents per working-age person.
Polymarket and Metaculus have a gap between them for 100k COVID cases before 2022 (95% vs 85%). Does that mean you could bet them against each other for a guaranteed payoff? If you can't, does that indicate a deficiency in their payoff algorithm? Or goes the gap persist because of fees or some other impediment?
> check a box saying you read some long contracts which you realistically did not read
Note that Kalshi's various agreements are not entirely screwing around. In particular, there are specific prohibitions on "participating in a contract if you ... have material non-public information regarding this contract, or ... have the ability to influence the outcome of the contract." I recommend assuming that a violation of this provision will be costly to you, whether or not it gets you in actual legal trouble, and generally treating this more like real securities trading, and less like play money cryptocurrency screwing-around, then one might otherwise be inclined.
> Right now it’s 50 across most of the world, except in Africa where it’s 80
Perhaps obvious in retrospect, but you might want to clarify that, even though this is called the "dependency ratio", it is in fact measured as a percentage, not really a ratio. I.e. that's 50% and 80% as many dependents as non-dependents.
Biden at 42 seems really cheap. I wouldn’t take 80 but 42? Anyone have reasons against that? Death tables are .05 probability of death per year at 80ish and that includes, I believe, nursing home vegetables and the like. Also Kamala winning a primary seems questionable.
Interestingly sportsbet.com.au has Harris at 2.75 (equivalent to 0.36 if I'm understanding the units correctly) and Biden at 3.25 (equiv 0.31).
Tempting to put down a hundred bucks on each, I don't see much chance of it being anyone else. Looking at the list of other low-odds candidates on sportsbet (Buttigeig, Sanders, AOC, O'Rourke, Yang) I just don't see anyone else realistic there.
Primaries are crowded often. Would Biden have been a good choice in the 16 primary? I can’t see Kamala with even odds to Biden really. Are moderates warming up to her? Even if they do I don’t see that.
I really do want to hear anyone’s case against say Biden at 70c / 2:1 win though, in case I’m wrong! I don’t use enough money to move anything don’t worry
This gives Biden and trump a 95% and 90% chance of surviving a four year term. His parents lived long - mother into 90s and father into 80s. They address health, conclude biden and trump are likely to remain healthy and mentally stable for a long time.
"There is no evidence available in the public record to indicate that either candidate is facing a major cognitive functioning challenge"
No medical evidence perhaps, but Joe Biden's has become a lot more incoherent with age hasn't he? For someone who knows more about cognitive decline, can something like that plateau for a few years?
We can distinguish general age related declines in energy and sharpness that happen to literally everybody, and cognitive decline in the severe sense. If you watch interviews in full and not cherry picked clips from a thousand hours of footage, he is firmly in the first camp. I’m not a neuroscientist though, but he seems as good as most people his age can be lol
Ok, but is that good enough to run for re-election when he's 82 or whatever? Normally, high IQ people are better than "most people their age".
There's a lot of politicians around the same age like Nancy Pelosi, Mitch McConnell and Bernie Sanders that as far as I know don't have hundreds of clips of them being incoherent.
number of media out of context clips is not a good measure of anything at all.
Nancy pelosi for instance has a few articles about her aids flagging serious cognitive decline, forgetting stuff, etc.
Biden has all those clips because a dozen paid gop guys and a few million republicans are scrounging the internet for anti Biden material. watching interviews he gives makes it clear he’s on the good end
He does have an extremely extremely mild stutter that comes out maybe ten times, and he says uh a lot and messes up a few words here and there in 1980. Not nearly as much stutter and word mistakes as now, but... I don’t think it’s that bad really.
.05 probability of death per year seems low. I’m not claiming to know the stats myself, but in 2019, men’s life expectancy was ~76 years. So if you’re already past that, using a probability that has an expected value of 20 years of life, that seems to be just too low, unless it ramps up super super fast at some point soon.
And yeah it does ramp up quickly after that. Death probability in general just slowly (but also quickly!) ramps up as you get older starting from the truly absurd age 10 p=0.000097.
It's not clear to me exactly what exact markets you're all talking about, but if they're "will be president on January 19th 2025" or something, you're tying your money up for over 3 years, which you should take into account.
Listing the prediction about "Haredi" population in Israel together with the estimates of total dependency makes a lot of sense, as this population is typically dependent: many of the Haredi men do not work or work in low paying teaching jobs.
The current level of dependency is barely sustainable; It's hard to imagine the current system persevering with 24% of the population.
Now at 19.1% for 2050 and 24.0% for peak. It looks like the initial estimates have dropped a lot. It would be interesting to have a related question on Haredi employment, which should rise.
Can prediction markets not change how prices map to probabilities to address the issue of fees skewing the prediction? Or is the concern more than fees drive down volume?
Thanks so much for banging on this drum. I had my own ideas for a while on 'how the fuck do you solve the problem of news being largely narrative nonsense', but the only solution i could think of was too complicated and seemed like it was ideal for a race of dispassionate computer-beings.
Prediction markets are much more plausibly a path out of 'the common knowledge is a mixture of data-driven accuracy, and bs promoted by the modern equivalent of priests enforcing an orthodoxy'.
Then those people will lose money over time, and they will give that money to people who are less wrong.
If there are negative consequences to believing in false narratives, people will do less of it.
Right now, there are social rewards (status, prestige) to believing in false narratives, and negative consequences (shame, ostracism, even losing your job) for deviating from false narratives.
"If there are negative consequences to believing in false narratives, people will do less of it."
People lose money betting on horse races all the time, but racecourses and the gambling industry are still operating. This kind of "pay real money" prediction market seems to me to be the same thing as betting "who will win the 2024 presidential election?" https://www.paddypower.com/politics?tab=usa
"Once everyone has been vaccinated or infected, it settles down into something about twice as bad as the flu."
I was surprised to read this. Partly because I don't think in the 2022-5 period everybody who is going to get infected will have already done so. But even allowing for that, it seems high - I was under the impression that in a population where most people (and very nearly all of the vulnerable) have been vaccinated, Covid will kill less people than flu rather than twice as many.
I think that's fair enough. Then my question follows Scott's - what proportion of deaths will be vaccinated/unvaccinated? And how precisely does this follow a country's vaccination rate?
According to the official UK data, fully vaccinated make up 54% of all deaths from delta. Israel’s data suggests a comparable ratio (though the sample size is still small).
According to Fauci, the vaccinated make up 0.8% of all recent deaths.
If you had to make a prediction - what is the probability that Fauci’s statement is true or close (say, up to a factor of 3) to reality?
75%? Maybe higher? There's a huge confounding factor between the US & the UK (not sure about Israel) which is that the UK has incredibly strong vaccination among the elderly (and surprisingly weak vaccination among the young) so the people most likely to die are disproportionately likely to also be vaccinated.
In the UK, about 92% of the people above 70 are fully vaccinated (many of them by the less effective AZ). In the US it's ~80%. This gives <2.5 factor, when we need >60 factor to explain the difference.
So if we have a 2.5 factor of vaccinations, plus lets say a 2-3 factor of effectiveness (no idea what the actual number is) plus your factor of 3 allowable error that gets us within a factor of 4. My prior that Fauci is either lying or wrong by a factor of 3 is probably 10%, my prior that there's another factor of 4 I didn't account for is probably somewhere in the 30% range, which gets me to roughly 75% confidence (this could be wrong, I didn't actually work through the Bayesian calculation).
I'm going to guess that you have a lower confidence in Fauci, but given my priors I think 75% confidence is still my guess.
In Israel (where Pfizer was used), the fully vaccinated made 58% (83/143) of recent hospitalizations and 75% (15/20) of recent deaths. Assuming this trend persists, would you adjust your estimate to under 50%?
> Dependency ratio is the ratio of non-working age people (eg elders, children) to working-age adults. Higher numbers mean more dependent people and greater economic burden. Right now it’s 50 across most of the world, except in Africa where it’s 80 (Africa has lots of kids!).
Nitpick: Assuming all of the dependants are kids, this means around 100 children per couple, which seems a bit high for a K-selected species? Realistically, I think there are some percentage signs missing here, e.g. the dependency ratio is really 0.5?
I have to admit I’d love to be an Analytical Storyteller, except that I’m functionally innumerate and shouldn’t be participating in prediction markets in any sense. But if someone fed me the numbers I’d love to do this kind of writing.
It'd be interesting to see if there are any arbitrage opportunities. Though these markets are probably too small and the population participating is too variable to develop reliable predictions about how poor prediction markets are at making predictions - that was a fun sentence. But if they consistently over or under estimate certain categories of topics, then one could make bets without really knowing anything about the question area.
I suppose part of the market element is to attract such arbitrage focused traders whose bets would then balance out the ratios....but with any low volume market, one could make significant money by 'providing' such services. Even a small 5-10% advantage would be great if one could find a consistent trend to under/over estimate health outcomes or narrow/broad topics of single states vs whole nations, etc.
I'm surprised and disappointed that "reddit predictions" isn't the thing I imagined when I heard that: Use a karmic prediction market to set initial scores on posts. This would actually solve a massive ongoing problem with reddit:
New posts rise or fall according to the whims of a very small sample of atypical voters. They can easily be sunk for stupid reasons that most members of the sub wouldn't agree with. Using a prediction market to set scores instead wouldn't have this problem.
Bettors would be answering a question like "what will the up/down ratio of this be once 150 total votes have been cast". Until the post reaches 150 votes, the post would be ranked according to their prediction instead than the actual vote ratio, which for the above reasons, at that stage, is too volitile to be used. Once it reaches 150 votes, the bettors receive their predictionkarma payouts and it's ranked according to the actual ratio from then on.
(note, reddit doesn't actually use vote ratio (`up/(up+down)`), it uses `up - down`. I've never come across a justification of this, it seems like that would amplify "oldest post wins" effects, so I wont advocate it, but if they'd prefer to keep that, the betting question could be adapted to "what will the score be after a day's viewing". Personally, that seems like a much more annoying question, because it's confounding itself with the number of people who happen to come online that day, which I'm going to be much less interested in learning to predict, the online system shouldn't really care about that.)
("what will the score be after a thousand views" would be less confounded, it I think it would be an even better measure of reception than up/(up+down) ratio, but view count might be easier to manipulate/spam than vote count. Reddit will have put a lot of effort into preventing vote manipulation, they will have had less of a reason to work on view manipulation, but it's conceivable that the solutions to vote manipulation can transfer to preventing view manipulation.)
The "fortified essays" seem quite interesting, and reading the description, it doesn't seem like it's a full-time job, so maybe I'll do it on the side. I could imagine writing things in the style of "X: much more than you wanted to know"
I’m writing to lodge a minor objection to the title of this series. I find it hard to believe I’m the only one who, every time one of these comes in, finds himself stuck with that horrible Bangles song from the 1980s stuck in my head for the rest of the day. (“Just another man[t]ic Monday.”)
No you are not the only one. I like the mantis pic a lot though.
Wait a minute… where is the preying mantis?
Pics for a post don't appear on the post itself for some reason - only next to links to the post.
Rename it Mantis Monday! Mantises are cool. No so much the Bangles. Lol!
Isn't that the point, though? :)
Right! "Will I make it through this day?"
I have the same problem and it's really annoying. I don't even hate the song that much; I just don't want it stuck in my head one day a week.
Well I didn't have it in my head until *now* - perils of being 80s kids?
For anyone too young to know the song—for the love of God, don’t look it up, at least not while this series continues! Much like Roko’s Basilisk, you will rue the weekly suffering you endure due to knowledge that you can’t unlearn!
Makes sense. Personally, it always makes me think of Dr. Mantis Toboggan, MD. "I got your test results! You're positive! Ya got the HIV! "
I already struggle with this problem whenever I have the thought "huh it's Monday" and Scott is certainly not helping. I wouldn't mind, except "wish it was Sunday / 'cause that's my fun day" has always grated for some reason.
They kind of walk like an egyptian too https://www.youtube.com/watch?v=NXN_3fgYcjM
https://youtu.be/Cv6tuzHUuuk?t=99
Bwahahaha, if you want an 80s pop earworm, I've got one for you! (Click on link at own risk) 😁
https://www.youtube.com/watch?v=FTQbiNvZqaY&list=PLCD0445C57F2B7F41&index=11
or https://www.youtube.com/watch?v=dQw4w9WgXcQ
You may just have to....Let It Go.
Ummm ... that song was actually written by ... Prince.
It's also, not by coincidence, really, REALLY good. Does it stick in your head? Yes. Prince was an amazing pop songwriter.
Knowing that it's actually Prince...
(who was smitten with Susanna Hoffs, but let the Bangles go ahead and record MM even though she [not unkindly] turned him down)
... go back and listen to the bridge and see if you now smack yourself in the forehead, that bridge is so clearly his genius...
This was during his absurd imperial phase, when his output was so great that even his table scraps could be top 10 hits and he launched bands seemingly every few months.
I like "Manic Monday"
B
This Magic parody might help ?
https://www.youtube.com/watch?v=azEg75bUHU4
The 9-12% range of estimate is a pretty large uncertainty for the baseline, 24% could be double or almost 3x.
Did you have to give an SSN for Kalshi? I know it's regulated, but that still bothers me a little.
I didn't have to when I signed up last week
Why should Kalshi be less than 10% on their fees? That’s what a typical sports book would be charging, and their motivations (i.e. profit) are the same. I guess a few online sports books go cheaper, but they have more competition. Frankly I’m surprised Kalshi’s fees are that low if they’re the first regulated company into this market.
Ignoring the issue of "should" (since you're right, it's competition driven), the comparison to sportsbooks isn't quite right. While sportsbooks do resemble a market at times as they try and balance their action, they can leave themselves substantially exposed to risk on one position or another.
Kalashi doesn't take on any risk however. They're operating an exchange without taking on any risk, so in the abstract, you'd expect their fees to be lower.
I agree that they should be lower! But I'm very confused bc I think they are lower. The PDF Scott linked shows the largest fee as $0.0175 for $0.50 of profit, so isn't that 3.5%?
I had the same impression when I did the signup. I presumed there was some nuance I was missing, but maybe not?
Why don’t stock exchanges charge 10% per transaction? They are for profit. Why shouldn’t credit card companies charge 10%?
It’s bad for the goal of the service and the actual benefit of the service. So both practically and (eventually) for profit, they should charge less.
About sports books see below
I think neither of these are fair comparisons, since both make their profits heavily through other avenues that aren’t open to Kalshi.
First off, a lot of modern companies charging 0% royalties make money in other ways, the primary ones seem to be matching you with other customers to cash in on spreads, selling the right to place your order (I don’t understand why this works) and micro-interest from when your money sits around between transactions. Assumedly since Kalshi is the market there aren’t spreads (so you’re saving money here), they place orders themselves, and their volume is low.
With credit cards, they do charge a percentage amount, but they charge it to the vendor (though I don’t think it’s 10%). But the way they make tons of money is off of the ~20% APR loans customers sometimes take out.
I looked it up, credit cards range from 1.5% to 3.3% which at the high end is comparable to Kalshi if you take the 3.5% rate the representative said it evened out to.
credit cards often give back between 1 and 2% in rewards and eat the cost of fraud/theft
My understanding is that this only holds for users with good credit. I’d had friends who were unable to get cash back cards (or even who had to get secured cards) because their credit was insufficient.
considering that credit scores come from how much (and old) lines of. credit you have, how much of it you use, and how reliably you pay bills, this all adds up to “how much do you actually need a credit card?”
Not having a credit card would literally not affect my day-to-day life. Thus they have to compete to convince a customer like me to actually get a card. So they give me rewards like cash back. For someone who actually needs the card, they’ll usually have worse credit and thus not get as good a card.
So in my view, what’s going on is that Kalshi hasn’t figured out a model for price discrimination yet. I expect if we ever have major traders on such a company, they’ll be able to negotiate reduced rates.
Having done some research on the payments industry, at least in the US, I think credit card fees are seldom higher than 2.0% these days (unless we're talking PayPal). And they've been trending consistently downward.
The loans are made by the banks that issue cards, and for the banks I think those are usually the most lucrative part of the credit card business, but the banks aren't the only players. The "networks" like Visa and Mastercard take the smallest part of the fee (I think usually less than 0.5%), but they're very profitable because they also do the least work. IIRC the payment processors (e.g. Global Payments) take the largest part of the fee, or perhaps they're roughly tied with the banks, but they haven't been as profitable because they're also out there doing the most work and they have the least market power.
All this is different for AMEX, as they're vertically integrated and own the whole process. But they're also much smaller.
UK betting exchanges mostly have 2% commission. Betdaq is 0% for most sports (presumably it's a loss leader).
Thanks for that info. The betting exchanges are a closer comparison to what Kalshi is doing than a sports book would be.
I notice there are no predictions about vaccine breakthrough, though I see there's one about variants.
There should be a "things that make vaccines fail" prediction. Turns out, steroids +/- 2 weeks from vaccination can lower the effectiveness. Since shots are being given by big-box stores and not updating to PCP charts, there is a likelihood of this type of thing. It may be contributing to me getting breakthrough covid, fortunately not a bad case, but annoying, because not supposed to be happening. How about, percent of breakthrough infections influenced by drug interactions w/vaccine. If I can figure out how to work it I'll put that up.
"percent of Americans who die of COVID between now and 1/1/22 who were fully vaccinated" is ambiguous, and I first read it as "X% of the set of Americans will have been vaccinated and also die from COVID" rather than "out of the set of Americans who die of COVID, X% will have been vaccinated". Thus my WTF reaction to your estimate of 2.5%.
I don't think the original wording is actually ambiguous, although I also misunderstood it on a first reading. (Probably because I was expecting a different question.)
Oh, wow, I had the same misreading. Probably because the number I am most interested in is "as a fully vaccinated person, what percent chance do I have of dying/hospitalization/long COVID?" so I just assumed that paragraph was about the question I had in mind.
(As for the answer to my question: all the articles seem to say "almost no chance of dying/hospitalization", which is reassuring, although as Scott mentions having an actual predicted number would be better. The articles I've seen are quiet on long COVID speculation though, besides a throwaway line about "concern among scientists" in https://www.nytimes.com/2021/07/22/health/coronavirus-breakthrough-infections-delta.html .)
Is there any long-term prediction markets covering ambiguous questions in the realm of science? I'd be particularly interested in what they would predict the field of nutrition science will say about e.g. what is the healthiest way to eat in 10, 20, 30 years. Will they have settled on low-carb? Back to low-fat? Etc.
I'd put a buck on non-obese
I'd be very surprised if there is any unambiguously "healthiest way to eat" for an arbitrary person: https://slimemoldtimemold.com/2021/07/07/a-chemical-hunger-part-i-mysteries/
That's really interesting, and fortunately questions around diet are not, it seems, connected to slime mould
Even if there was, I'd be very surprised if nutrition scientists figured out what that was within the next 30 years.
Or that, even if they (or even one of them) did, that anyone else would be able to spot the signal in all the noise that is nutrition 'research'!
There are some questions about nutrition on Metaculus: Keto[1], Caloric restriction[2] and Soylent[3].
[1]: https://www.metaculus.com/questions/3739/keto
[2]: https://www.metaculus.com/questions/4094/caloric
[3]: https://www.metaculus.com/questions/6967/soylent
China dependency ratio - I guess that China will raise the retirement age and that will slow the growth of the dependency ratio
Yes, a totalitarian system like China can raise the retirement age a bit more easily than Germany.
Though even for dictatorships, cutting social welfare is one of the hardest things to do.
The cutoffs for the dependancy ratio calculation are fixed at 14 and 64. You can't fiddle with the numbers by changing the retirement age.
Doesn't that just mean changes to the retirement age above 64 won't affect these numbers? The retirement age for most people in China is currently below 64 (at least officially - I don't have any idea how common it is for people to work beyond the official retirement age). Sorry if I still don't understand how this actually works.
No, it won't affect the numbers. Lots of developing countries don't even have retirement ages, so a fixed cutoff is the simplest way to quantify who is too old or too young to work.
China's median age is around 37, so their demographic crisis won't hit till after 2030 (around 2040 I guess) when the effects of the one-child policy will become obvious in the workforce. Unless they open to mass immigration of course.
China is pushing now for women to have more children - they have just instituted their Third Child Policy. So between social pressure (and China already has its social credit ratings system in place) to stop being a career woman, get married, and start a family, and already married people who might want more kids, presumably the estimation here is that China will be getting a lot of new babies over the years.
https://en.wikipedia.org/wiki/Three-child_policy
I have also read online some rather ugly speculation that this is also part of the treatment of the Ugyhurs: move in Han men, get them married (consent not being too important) to Ugyhur women, and start producing bouncing Han Chinese babies which is two birds with one stone about propping up the birth rate and 'assimilating' the Ughyur minority into the mainstream. How true this is I have no idea, but I can't dismiss it out of hand.
https://en.wikipedia.org/wiki/Uyghur_genocide
The cutoff is in fact 19- and 65+ for the metaculus question. So every child that is born this year or later will *increase* the dependency ratio in 2039, not *decrease*. This is the wrong direction.
I'm skeptical that raising the allowed number of children to 3 will have that much impact on the birth rate. I don't think there was that much of a bump when they went from 1 to 2. There's already a lot of social pressure in China to have a "traditional family" but the default is 1 kid and not many people actually want / can afford more.
China is certainly trying to find ways to alleviate the pressures that deter people from having more kids - that's why they're cracking down on these after school tutoring companies right now. It just doesn't seem to be very easy to get people to have more kids though (see for example Singapore).
This is right. Raising fertility in modern societies is extremely difficult, the anti-natal pressures within modernity are immense, and they seem to affect NE Asian societies particularly acutely. China maybe has a chance of pushing back, but they haven't done anything yet to appear very serious on this topic.
"Appearing serious" would probably involve things like promoting anti-natal media and banning all media that so much as hints at glamorizing childlessness or small families, and requiring 3+ children (or even more) for promotion within the CCP.
I also think that if these things worked, it might still take decades. Pro-natal propaganda probably won't have much effect on people who are 25 today, but 20 years of it might affect people who are 5 today.
Scott, where did you get the 10% number from for the Kalshi fees? The document you linked shows the largest fee as $0.0175 for $0.50 of profit, so isn't that 3.5%?
So I'm trying to see what markets are on Kalshi but I can't find a page for it (without signing up, anyway -- I just want to see all the markets, I don't want to actually sign up). Is there a way to see this? I can find pages for individual markets, but not for the overall list of markets. (It isn't https://kalshi.com/markets/ ; that doesn't work.)
One ought to be able to see the list of markets without signing up!
Idt you can. I saw in a discord that there's this link which shows all the regulatory docs for the markets: https://kalshi.com/regulatory/product-filings
That's certainly interesting, although obviously not what I was looking for...
...anyway, they ought to fix that.
Agreed
I had a similar problem - perhaps this will help you: https://metaculusextras.com/kalshi
Or https://metaforecast.org/?query=Kalshi is even better
Am I misunderstanding or are the fees for Kalshi quite a bit less than 10%?
It looks like for 1-cent contracts the fee is 7%, but it looks like it is lower than that for higher-cost contracts - e.g. the fee for one hundred 50-cent contracts is $1.75, which is only 3.5%.
And for one hundred 99-cent contracts, the fee is 7 cents, which is less than 0.1%.
So on average the fee is around 3.5%? I might be misunderstanding the fee schedule though.
This is the exact understanding I had as well. My math also came out to 3.5%
They also say that:
> We charge this fee only to “liquidity takers,” who place orders that match orders sitting on the orderbook to complete a trade. We do not charge a fee to “liquidity makers” who place an order that rests on the orderbook waiting to be matched.
So it seems the fee is only for one side of a trade. So on average your fee would be half of that?
I've been trading on Polymarket for two months, and it's been incredible. I am learning a lot more on the subjects I am wagering on, and it is still vastly inefficient. I have seen some people do very little work to guarantee themselves 5 figure payouts. This is what trading bonds in the 80s much have felt like to an extent.
I should note that I just put $100 into this and I had to pay for almost $30 of Etherium on top of the USDC to get my $100 in, so be aware. (My $100 was $99.64 after some fees but that's the least of it.)
All these prediction markets, and not a single one where I can bet on who the final Smash character will be. :P
I had not heard of dependency ratios before, and I was VERY surprised at those numbers until I looked "dependency ratio" up on Wikipedia and discovered that it is conventionally multiplied by 100 (i.e. expressed as a percentage). So a dependency ratio of "50" means 0.5 dependents per working-age person, not 50 dependents per working-age person.
Oh! That makes a lot more sense now...
Odd. Why not just state the rate?
Polymarket and Metaculus have a gap between them for 100k COVID cases before 2022 (95% vs 85%). Does that mean you could bet them against each other for a guaranteed payoff? If you can't, does that indicate a deficiency in their payoff algorithm? Or goes the gap persist because of fees or some other impediment?
Metaculus points aren’t dollars!
> check a box saying you read some long contracts which you realistically did not read
Note that Kalshi's various agreements are not entirely screwing around. In particular, there are specific prohibitions on "participating in a contract if you ... have material non-public information regarding this contract, or ... have the ability to influence the outcome of the contract." I recommend assuming that a violation of this provision will be costly to you, whether or not it gets you in actual legal trouble, and generally treating this more like real securities trading, and less like play money cryptocurrency screwing-around, then one might otherwise be inclined.
> Right now it’s 50 across most of the world, except in Africa where it’s 80
Perhaps obvious in retrospect, but you might want to clarify that, even though this is called the "dependency ratio", it is in fact measured as a percentage, not really a ratio. I.e. that's 50% and 80% as many dependents as non-dependents.
Biden at 42 seems really cheap. I wouldn’t take 80 but 42? Anyone have reasons against that? Death tables are .05 probability of death per year at 80ish and that includes, I believe, nursing home vegetables and the like. Also Kamala winning a primary seems questionable.
Interestingly sportsbet.com.au has Harris at 2.75 (equivalent to 0.36 if I'm understanding the units correctly) and Biden at 3.25 (equiv 0.31).
Tempting to put down a hundred bucks on each, I don't see much chance of it being anyone else. Looking at the list of other low-odds candidates on sportsbet (Buttigeig, Sanders, AOC, O'Rourke, Yang) I just don't see anyone else realistic there.
Primaries are crowded often. Would Biden have been a good choice in the 16 primary? I can’t see Kamala with even odds to Biden really. Are moderates warming up to her? Even if they do I don’t see that.
I really do want to hear anyone’s case against say Biden at 70c / 2:1 win though, in case I’m wrong! I don’t use enough money to move anything don’t worry
Health more than death, presumably?
https://www.icaa.cc/media/presidential_lifespan_and_healthspan-draft_for_release_1.pdf
This gives Biden and trump a 95% and 90% chance of surviving a four year term. His parents lived long - mother into 90s and father into 80s. They address health, conclude biden and trump are likely to remain healthy and mentally stable for a long time.
"There is no evidence available in the public record to indicate that either candidate is facing a major cognitive functioning challenge"
No medical evidence perhaps, but Joe Biden's has become a lot more incoherent with age hasn't he? For someone who knows more about cognitive decline, can something like that plateau for a few years?
We can distinguish general age related declines in energy and sharpness that happen to literally everybody, and cognitive decline in the severe sense. If you watch interviews in full and not cherry picked clips from a thousand hours of footage, he is firmly in the first camp. I’m not a neuroscientist though, but he seems as good as most people his age can be lol
major cognitive functioning challenge isn’t just “general decline”, it’s major, as in dementia
(Dementia as an example. I wish substack had edit button.)
"he seems as good as most people his age can be"
Ok, but is that good enough to run for re-election when he's 82 or whatever? Normally, high IQ people are better than "most people their age".
There's a lot of politicians around the same age like Nancy Pelosi, Mitch McConnell and Bernie Sanders that as far as I know don't have hundreds of clips of them being incoherent.
number of media out of context clips is not a good measure of anything at all.
Nancy pelosi for instance has a few articles about her aids flagging serious cognitive decline, forgetting stuff, etc.
Biden has all those clips because a dozen paid gop guys and a few million republicans are scrounging the internet for anti Biden material. watching interviews he gives makes it clear he’s on the good end
Lmao he was randomly insulting people in 1988 https://m.youtube.com/watch?v=D1j0FS0Z6ho
https://m.youtube.com/watch?v=lueMrnGpmK8
He does have an extremely extremely mild stutter that comes out maybe ten times, and he says uh a lot and messes up a few words here and there in 1980. Not nearly as much stutter and word mistakes as now, but... I don’t think it’s that bad really.
.05 probability of death per year seems low. I’m not claiming to know the stats myself, but in 2019, men’s life expectancy was ~76 years. So if you’re already past that, using a probability that has an expected value of 20 years of life, that seems to be just too low, unless it ramps up super super fast at some point soon.
https://www.ssa.gov/oact/STATS/table4c6.html
0.05 is at age 78, so pretty darn close.
I see, thanks for providing the source.
And yeah it does ramp up quickly after that. Death probability in general just slowly (but also quickly!) ramps up as you get older starting from the truly absurd age 10 p=0.000097.
It's not clear to me exactly what exact markets you're all talking about, but if they're "will be president on January 19th 2025" or something, you're tying your money up for over 3 years, which you should take into account.
The 2024 dem primary market, but yes. I had read in a past thread that this is why betting markets should be denominated in SPY.
Just signal boosting my Twitter account (https://twitter.com/metaculusextras) which shares the best comments and predictions from Metaculus each week
Scott you better apply for (contributing to) that ‘fortified essay’ stuff!
Listing the prediction about "Haredi" population in Israel together with the estimates of total dependency makes a lot of sense, as this population is typically dependent: many of the Haredi men do not work or work in low paying teaching jobs.
The current level of dependency is barely sustainable; It's hard to imagine the current system persevering with 24% of the population.
The prediction https://www.metaculus.com/questions/7513/-israeli-population-that-is-haredi-in-2050/ is down to 19.9% as of writing. The prediction for the peak https://www.metaculus.com/questions/7571/haredi-share-of-israel-at-peak/ is 24.4% which is the same as the 2050 estimate in Scott's post.
Anyway, Haredi employment rates will definitely increase. The current Israeli government is fortunately fixing that right now.
https://www.metaculus.com/questions/7571/haredi-share-of-israel-at-peak/
Prediction for the peak is only 24.1% now. Metaculus does not think that the Haredim will take over Israel.
Now at 19.1% for 2050 and 24.0% for peak. It looks like the initial estimates have dropped a lot. It would be interesting to have a related question on Haredi employment, which should rise.
20.0% for 2050 and 23.3% for peak. I should stop providing these updates though.
Can prediction markets not change how prices map to probabilities to address the issue of fees skewing the prediction? Or is the concern more than fees drive down volume?
Maybe it's not just a mapping but error bars that make it impossible to differentiate between 9% or 1%.
Thanks so much for banging on this drum. I had my own ideas for a while on 'how the fuck do you solve the problem of news being largely narrative nonsense', but the only solution i could think of was too complicated and seemed like it was ideal for a race of dispassionate computer-beings.
Prediction markets are much more plausibly a path out of 'the common knowledge is a mixture of data-driven accuracy, and bs promoted by the modern equivalent of priests enforcing an orthodoxy'.
Only if you assume people do not make decisions in prediction markets based on narratives.
Then those people will lose money over time, and they will give that money to people who are less wrong.
If there are negative consequences to believing in false narratives, people will do less of it.
Right now, there are social rewards (status, prestige) to believing in false narratives, and negative consequences (shame, ostracism, even losing your job) for deviating from false narratives.
But people make decisions based on false narratives and receive negative consequences all the time. How do prediction markets change this?
"If there are negative consequences to believing in false narratives, people will do less of it."
People lose money betting on horse races all the time, but racecourses and the gambling industry are still operating. This kind of "pay real money" prediction market seems to me to be the same thing as betting "who will win the 2024 presidential election?" https://www.paddypower.com/politics?tab=usa
"Once everyone has been vaccinated or infected, it settles down into something about twice as bad as the flu."
I was surprised to read this. Partly because I don't think in the 2022-5 period everybody who is going to get infected will have already done so. But even allowing for that, it seems high - I was under the impression that in a population where most people (and very nearly all of the vulnerable) have been vaccinated, Covid will kill less people than flu rather than twice as many.
I’m guessing that most predictors are assuming that a few years out, COVID vaccinations will drop to where flu vaccinations are (generally ~45% or so)
I think that's fair enough. Then my question follows Scott's - what proportion of deaths will be vaccinated/unvaccinated? And how precisely does this follow a country's vaccination rate?
"I think for the percent question I would guess something like 2.5% for this - right now it’s 1%"
--------------------------------------------------------------------
According to the official UK data, fully vaccinated make up 54% of all deaths from delta. Israel’s data suggests a comparable ratio (though the sample size is still small).
According to Fauci, the vaccinated make up 0.8% of all recent deaths.
If you had to make a prediction - what is the probability that Fauci’s statement is true or close (say, up to a factor of 3) to reality?
75%? Maybe higher? There's a huge confounding factor between the US & the UK (not sure about Israel) which is that the UK has incredibly strong vaccination among the elderly (and surprisingly weak vaccination among the young) so the people most likely to die are disproportionately likely to also be vaccinated.
In the UK, about 92% of the people above 70 are fully vaccinated (many of them by the less effective AZ). In the US it's ~80%. This gives <2.5 factor, when we need >60 factor to explain the difference.
So if we have a 2.5 factor of vaccinations, plus lets say a 2-3 factor of effectiveness (no idea what the actual number is) plus your factor of 3 allowable error that gets us within a factor of 4. My prior that Fauci is either lying or wrong by a factor of 3 is probably 10%, my prior that there's another factor of 4 I didn't account for is probably somewhere in the 30% range, which gets me to roughly 75% confidence (this could be wrong, I didn't actually work through the Bayesian calculation).
I'm going to guess that you have a lower confidence in Fauci, but given my priors I think 75% confidence is still my guess.
In Israel (where Pfizer was used), the fully vaccinated made 58% (83/143) of recent hospitalizations and 75% (15/20) of recent deaths. Assuming this trend persists, would you adjust your estimate to under 50%?
> Dependency ratio is the ratio of non-working age people (eg elders, children) to working-age adults. Higher numbers mean more dependent people and greater economic burden. Right now it’s 50 across most of the world, except in Africa where it’s 80 (Africa has lots of kids!).
Nitpick: Assuming all of the dependants are kids, this means around 100 children per couple, which seems a bit high for a K-selected species? Realistically, I think there are some percentage signs missing here, e.g. the dependency ratio is really 0.5?
I have to admit I’d love to be an Analytical Storyteller, except that I’m functionally innumerate and shouldn’t be participating in prediction markets in any sense. But if someone fed me the numbers I’d love to do this kind of writing.
It'd be interesting to see if there are any arbitrage opportunities. Though these markets are probably too small and the population participating is too variable to develop reliable predictions about how poor prediction markets are at making predictions - that was a fun sentence. But if they consistently over or under estimate certain categories of topics, then one could make bets without really knowing anything about the question area.
I suppose part of the market element is to attract such arbitrage focused traders whose bets would then balance out the ratios....but with any low volume market, one could make significant money by 'providing' such services. Even a small 5-10% advantage would be great if one could find a consistent trend to under/over estimate health outcomes or narrow/broad topics of single states vs whole nations, etc.
I'm surprised and disappointed that "reddit predictions" isn't the thing I imagined when I heard that: Use a karmic prediction market to set initial scores on posts. This would actually solve a massive ongoing problem with reddit:
New posts rise or fall according to the whims of a very small sample of atypical voters. They can easily be sunk for stupid reasons that most members of the sub wouldn't agree with. Using a prediction market to set scores instead wouldn't have this problem.
Bettors would be answering a question like "what will the up/down ratio of this be once 150 total votes have been cast". Until the post reaches 150 votes, the post would be ranked according to their prediction instead than the actual vote ratio, which for the above reasons, at that stage, is too volitile to be used. Once it reaches 150 votes, the bettors receive their predictionkarma payouts and it's ranked according to the actual ratio from then on.
(note, reddit doesn't actually use vote ratio (`up/(up+down)`), it uses `up - down`. I've never come across a justification of this, it seems like that would amplify "oldest post wins" effects, so I wont advocate it, but if they'd prefer to keep that, the betting question could be adapted to "what will the score be after a day's viewing". Personally, that seems like a much more annoying question, because it's confounding itself with the number of people who happen to come online that day, which I'm going to be much less interested in learning to predict, the online system shouldn't really care about that.)
("what will the score be after a thousand views" would be less confounded, it I think it would be an even better measure of reception than up/(up+down) ratio, but view count might be easier to manipulate/spam than vote count. Reddit will have put a lot of effort into preventing vote manipulation, they will have had less of a reason to work on view manipulation, but it's conceivable that the solutions to vote manipulation can transfer to preventing view manipulation.)